MTA Problem in High‑Precision LiDAR Data and Its Correction Algorithms

High‑precision acquisition vehicles integrate laser ranging, high‑performance inertial navigation, and high‑resolution cameras into a mobile mapping system. The vehicles developed by the Gaode high‑precision team feature high accuracy, fast speed, short data generation cycles, high automation, safety, and large information volume.

To ensure the accuracy of high‑precision map production, the vehicles use state‑of‑the‑art laser rangefinders with long measurement distance and high point‑cloud density, achieving a scanning frequency of up to one million points per second.

The high scanning frequency introduces a specific noise called MTA (Multi‑Time‑Around). MTA occurs when a laser pulse intended for a distant object is mistakenly assigned to a nearer range, causing far‑away structures (e.g., buildings) to appear as noise on the road surface.

MTA severely hampers downstream data processing, automatic recognition, and map generation, leading to errors in both automated and manual workflows.

The article explains the principle of MTA, the internal laser mechanisms that address it, and the data‑processing algorithms used to correct it.

Figure 1: Example of MTA‑affected data.

MTA Principle

MTA originates from the laser ranging principle. Typical lidar scanners use the Time‑of‑Flight (TOF) method: a laser pulse is emitted at regular intervals, and the time difference between emission and reception is used to calculate distance.

When the maximum detectable distance (Dmax) exceeds the laser pulse interval distance (Dpulse), multiple pulses can coexist in the air. The order of received pulses no longer matches the emission order, causing the receiver to miscalculate the TOF and produce MTA errors.

Figure 2: MTA interval illustration. For the Gaode vehicle, Dpulse is 150 m, so objects farther than 150 m may suffer from MTA.

Laser Internal Mechanisms to Mitigate MTA

Manufacturers employ two main strategies:

Neighbourhood continuity assumption: most real‑world surfaces (roads, signs, buildings) are locally continuous, so adjacent laser points should have similar distances.

Variable‑period measurement technique: the interval between successive laser pulses is varied in a periodic pattern. When a point is shifted to an incorrect MTA interval, its distance series becomes non‑continuous, which can be detected.

Figure 4: Variable‑period emission.

Figure 5: Periodic variation parameters.

Figure 6: Incorrect MTA interval causing layered point‑cloud artifacts.

MTA Correction Algorithm

The algorithm consists of four main steps:

Neighbourhood construction: because lidar scans in circular rings, both temporal adjacency within a ring and spatial adjacency across neighboring rings are considered.

Weighted statistical analysis: for each point, compute distance and reflectivity continuity weights (inverse variance). Larger weights indicate the correct MTA interval.

Performance optimization: limit the search to the two most likely MTA intervals (MTA1 and MTA2) based on the sensor’s 300 m detection range, and apply multithreaded processing.

Algorithmic refinements: discard ground points using vehicle height, use only the first return for multi‑echo points, determine spatial adjacency via scan angle and range, and treat isolated points with poor continuity as outliers.

Figure 7: Neighborhood search.

Figure 8: Selected weighting functions (Gaussian).

Experimental results show that the algorithm effectively restores points shifted by MTA, as illustrated in Figures 9 and 10.

Figure 9: Before and after MTA correction (unprocessed vs. processed).

Figure 10: Restored point‑cloud after MTA correction.

Summary and Outlook

The MTA correction algorithm is a critical component of the high‑precision map data processing pipeline; without it, automation is impossible. The algorithm also reduces dependence on proprietary vendor software, saving significant costs.

Attempts to apply machine‑learning methods (SVM, Random Forest) faced challenges such as sample imbalance and prohibitive computational load for billion‑point blocks. Consequently, a custom evaluation framework aligned with production requirements was devised, enabling reliable and rapid algorithm assessment.

Currently, the MTA correction algorithm is deployed in production, processing tens of thousands of kilometers of point‑cloud data with stable performance.